Providing adversarial perturbations to media

Abstract

A method for applying perturbations to electronic media is described. A computing device may receive an electronic request to upload an electronic media to an internet-based service and perform a pre-processing operation on the electronic media based on the electronic request and a feature of the electronic media. In some examples, the computing device may perform a security action on the electronic media. For example, the computing device may apply perturbations to elements of the electronic media based on a gradient of a model. The computing device may transmit the electronic media to the internet-based service.

What is claimed is: 
     
       1. A method for applying perturbations to electronic media, the method being performed by one or more computing devices comprising at least one processor, the method comprising:
 receiving an electronic request to upload an electronic media to an internet-based service; 
 performing, by the one or more computing devices, a pre-processing operation on the electronic media based at least in part on the electronic request and a feature of the electronic media, wherein the pre-processing operation comprises securing a privacy of the electronic media feature from a machine learning-based recognition algorithm associated with the internet-based service by applying a perturbation to the electronic media feature according to a gradient of the machine learning-based recognition algorithm; and 
 transmitting, by the one or more computing devices, the electronic media to the internet-based service. 
 
     
     
       2. The method of  claim 1 , wherein the pre-processing operation is performed prior to the transmitting of the electronic media to the internet-based service to upload the electronic media to the internet-based service. 
     
     
       3. The method of  claim 1 , wherein performing the pre-processing operation comprises:
 performing an analysis on the electronic media; 
 detecting facial data associated with the electronic media based at least in part on the analysis; and 
 performing a security action on the electronic media based at least in part on the detected facial data. 
 
     
     
       4. The method of  claim 3 , wherein performing the security action comprises:
 applying perturbations to elements of the electronic media based at least in part on a gradient of a model. 
 
     
     
       5. The method of  claim 4 , wherein the electronic media comprises a digital image and the elements comprises a plurality of pixels of the digital image, and wherein applying the perturbations comprises:
 applying a perturbation to each pixel of the digital image based at least in part on an algorithm. 
 
     
     
       6. The method of  claim 4 , wherein the electronic media comprises a digital image and the elements comprises a plurality of pixels of the digital image, and wherein applying the perturbations comprises:
 applying a perturbation to each pixel associated with the facial data of the digital image. 
 
     
     
       7. The method of  claim 6 , wherein an identity associated with an individual of the facial data is masked to one or more image recognition algorithms associated with the internet-based service based at least in part on the applied perturbations to the facial data of the digital image. 
     
     
       8. The method of  claim 4 , wherein the model of the gradient is associated with a neural network model using a public database of electronic media. 
     
     
       9. The method of  claim 1 , wherein performing the pre-processing operation comprises:
 identifying metadata associated with the electronic media; and 
 removing the metadata from the electronic media based at least in part on the identifying. 
 
     
     
       10. The method of  claim 1 , wherein the feature of the electronic media comprises facial data. 
     
     
       11. The method of  claim 1 , wherein the electronic media comprises at least one photograph image or at least one video image, or both. 
     
     
       12. The method of  claim 1 , wherein the one or more computing devices comprise at least one of a network proxy, or a server, or a mobile phone running an application, or any combination thereof. 
     
     
       13. The method of  claim 1 , wherein performing the pre-processing operation on the electronic media is performed by a network proxy. 
     
     
       14. A computing device configured for applying perturbations to electronic media, further comprising:
 one or more processors; 
 memory in electronic communication with the one or more processors, wherein the memory stores computer executable instructions that when executed by the one or more processors cause the one or more processors to: 
 receive an electronic request to upload an electronic media to an internet-based service; 
 perform a pre-processing operation on the electronic media based at least in part on the electronic request and a feature of the electronic media, wherein the pre-processing operation comprises securing a privacy of the electronic media feature from a machine learning-based recognition algorithm associated with the internet-based service by applying a perturbation to the electronic media feature according to a gradient of the machine learning-based recognition algorithm; and 
 transmit the electronic media to the internet-based service. 
 
     
     
       15. The computing device of  claim 14 , wherein the pre-processing operation is performed prior to the transmitting of the electronic media to the internet-based service to upload the electronic media to the internet-based service. 
     
     
       16. The computing device of  claim 14 , wherein the instructions for performing the pre-processing operation are further executable by the one or more processors to:
 perform an analysis on the electronic media; 
 detect facial data associated with the electronic media based at least in part on the analysis; and 
 perform a security action on the electronic media based at least in part on the detected facial data. 
 
     
     
       17. The computing device of  claim 16 , wherein the instructions for performing the security action are further executable by the one or more processors to:
 apply perturbations to elements of the electronic media based at least in part on a gradient of a model. 
 
     
     
       18. A non-transitory computer-readable medium storing computer executable instructions that when executed by the one or more processors cause the one or more processors to:
 receive an electronic request to upload an electronic media to an internet-based service; 
 perform a pre-processing operation on the electronic media based at least in part on the electronic request and a feature of the electronic media, wherein the pre-processing operation comprises securing a privacy of the electronic media feature from a machine learning-based recognition algorithm associated with the internet-based service by applying a perturbation to the electronic media feature according to a gradient of the machine learning-based recognition algorithm; and 
 transmit the electronic media to the internet-based service. 
 
     
     
       19. The non-transitory computer-readable medium of  claim 18 , wherein the pre-processing operation is performed prior to the transmitting of the electronic media to the internet-based service to upload the electronic media to the internet-based service. 
     
     
       20. The non-transitory computer-readable medium of  claim 18 , wherein the feature of the electronic media comprises facial data.